CN107529222A - A kind of WiFi indoor locating systems based on deep learning - Google Patents
A kind of WiFi indoor locating systems based on deep learning Download PDFInfo
- Publication number
- CN107529222A CN107529222A CN201710833040.8A CN201710833040A CN107529222A CN 107529222 A CN107529222 A CN 107529222A CN 201710833040 A CN201710833040 A CN 201710833040A CN 107529222 A CN107529222 A CN 107529222A
- Authority
- CN
- China
- Prior art keywords
- module
- signal strength
- fingerprint
- data
- characteristic fingerprint
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
- G01C21/206—Instruments for performing navigational calculations specially adapted for indoor navigation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/0252—Radio frequency fingerprinting
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/10—Position of receiver fixed by co-ordinating a plurality of position lines defined by path-difference measurements, e.g. omega or decca systems
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W84/00—Network topologies
- H04W84/02—Hierarchically pre-organised networks, e.g. paging networks, cellular networks, WLAN [Wireless Local Area Network] or WLL [Wireless Local Loop]
- H04W84/10—Small scale networks; Flat hierarchical networks
- H04W84/12—WLAN [Wireless Local Area Networks]
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Automation & Control Theory (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The invention belongs to indoor positioning technologies field, more particularly, to a kind of WiFi indoor locating systems based on deep learning.Off-line data acquisition module (100), the thick fingerprint base that it includes being sequentially connected establish module (200), characteristic fingerprint storehouse extraction module (300), online data Fusion Module (400) and target location output module (500).The system solves indoor reception signal strength signal in space-time threshold, the signal fluctuation sex chromosome mosaicism caused by multipath effect, signal fadeout and other noise jammings, pass through the environment attribute inside depth confidence web inquiry signal, extract characteristic fingerprint and carry out final target positioning, and effectively achieve the inaccessiable positioning precision of current location technology.
Description
Technical field
The invention belongs to indoor positioning technologies field, more particularly, to fixed in a kind of WiFi rooms based on deep learning
Position system.
Background technology
It is applied to indoor environment more and more widely with the service of position, the research for indoor positioning technologies attracts
Increasing scientific research and commercial pursuit person.Indoor positioning technologies based on WiFi fingerprints turn into most popular indoor positioning skill
One of art, in various indoor place input applications.However, the fluctuation of wireless signal inherently is in positioning rank
Position error very big Duan Zaocheng.In addition, in the indoor environment of complexity, due to multipath effect, signal fadeout and other noises
Influence to receive signal intensity complicated change in space-time threshold.
In order to solve the technical barrier of the above, there are many technical staff to be calculated by classical KNN algorithms, machine learning
The positioning precision of the indoor positioning technologies of method and the lifting of some deep learning networks based on WiFi fingerprints.But in targeting scheme
Universality and final positioning precision in terms of following defect be present：
(1) in the existing indoor positioning technologies based on WiFi fingerprints, traditional location algorithm can not meet high-precision fixed
The requirement of position.In position fixing process, because the fluctuation of signal is too big, positioning result is caused easily to drift about, individual other positioning misses
The accuracy requirement of difference significantly larger than positioning.
(2) in the technical scheme of WiFi fingerprints is handled with machine learning, due to being needed using machine learning processing data
The data operation of high complexity, substantial amounts of calculation resources are taken in a mobile device, influence the application of mobile device other software,
And the endurance of mobile device can be substantially reduced.
(3) deep learning is introduced into the indoor positioning technologies of WiFi fingerprints, it is necessary to which gathering substantial amounts of data carries out fingerprint
Training, this consumes substantial amounts of human resources in offline link, improves technology application cost.In addition, often most of depth
A large amount of channel status information signals are trained in learning method selection.However, the signal can not be acquired and position in mobile phone terminal, greatly
The big universality for reducing the program.
The content of the invention
Part in view of the shortcomings of the prior art, the present invention propose a kind of WiFi indoor positionings based on deep learning
System, to solve indoor reception signal strength signal in space-time threshold, due to multipath effect, signal fadeout and other noise jammings
Caused signal fluctuation sex chromosome mosaicism, by the environment attribute inside depth confidence web inquiry signal, extraction characteristic fingerprint is carried out
Final target positioning, and effectively achieve the inaccessiable positioning precision of current location technology.
The present invention adopts the following technical scheme that：
A kind of WiFi indoor locating systems based on deep learning, it includes the off-line data acquisition module being sequentially connected
(100), thick fingerprint base establishes module (200), characteristic fingerprint storehouse extraction module (300), online data Fusion Module (400) and mesh
Cursor position output module (500), wherein,
The off-line data acquisition module (100) is used for the physical address information for obtaining access point, and collection is located at reference point
Locate offline received signal strength data, and these data are transferred to thick fingerprint base and establish module (200)；
The offline received signal strength data that the thick fingerprint base is established at all reference points of module (200) traversal so that
Offline received signal strength and the position coordinates of reference point correspond and generate thick fingerprint base；
Receive signal strength data offline at characteristic fingerprint storehouse extraction module (300) normalization reference point, its is defeated
Enter to four layer depth confidence networks and be trained, four layer depth confidence networks output weight and bias matrix storage are characterized finger
Line；
The online data Fusion Module (400) gathers received signal strength data and it is normalized place in real time
Reason, recycles the characteristic fingerprint at each reference point that the received signal strength data gathered in real time is reconstructed, by reconstruct
Receive former data input of the signal strength data with gathering in real time to RBF, calculate diversity factor between the two, be somebody's turn to do
Diversity factor characterizes the probability for receiving signal strength data and appearing in the reference point gathered in real time, travels through all reference points,
Probability of the current received signal strength data at each reference point is calculated, finally by the weight meter of reference point locations
Calculate geographical position corresponding to real-time reception signal strength data；
Described target location output module (500) carries out the target location estimated based on data anastomosing algorithm defeated
Go out, complete the positioning to target location.
The technical program further optimizes, and the off-line data acquisition module (100) includes wireless sensor module
(101) and offline received signal strength acquisition module (102), wireless sensor module (101) be used to scanning all access points without
The physical address of line sensor, offline received signal strength acquisition module (102), which is used to gather to come from reference point, to be connect
The offline of access point wireless senser receives signal strength data.
The technical program further optimizes, and the thick fingerprint base is established module (200) and recorded including reference point locations
Module (201), thick fingerprint base generation module (202) and thick fingerprint base output module (203)；Thick fingerprint base establishes module (200)
To be gathered by the offline received signal strength acquisition module (102) of reference point reference point it is offline receive signal strength data with by
Position coordinates at reference point locations logging modle (201) record reference point travels through all reference point shapes with being corresponded
Into thick fingerprint base, thick fingerprint base output module (203) is delivered to.
The technical program further optimizes, and characteristic fingerprint storehouse extraction module (300) includes offline reception signal
It is defeated to normalize module (301), four layer depth confidence mixed-media network modules mixed-medias (302), characteristic fingerprint extraction module (303) and characteristic fingerprint storehouse
Go out module (304)；It is offline to receive signal normalization module (301) and establish module (200) with thick fingerprint base to be connected, will each it refer to
The offline receiving intensity signal normalization at point place is between (0,1), by normalized received signal strength signal offline at four layers
It is trained in depth confidence mixed-media network modules mixed-media (302), characteristic fingerprint extraction module (303) is by four layer depth confidence mixed-media network modules mixed-medias
(302) weight in extracts with bias matrix to be stored as characteristic fingerprint, characteristic fingerprint storehouse output module (304) group
The formation characteristic fingerprint of the characteristic fingerprint at all reference points storehouse is closed to be exported.
The technical program further optimizes, and the online data Fusion Module (400) includes real-time reception signal intensity
Data acquisition module (401), real-time reception signal strength data normalization module (402), online data Fusion Module (403),
Read in module (404) and target location computing module (405) in reference point geographical position；The real-time reception signal intensity gathers mould
Block (401) gathers the real-time reception signal intensity of access point wireless senser, and real-time reception signal normalization module (402) will be real
When signal normalization between (0,1), online data Fusion Module (403) is with characteristic fingerprint storehouse extraction module (300) and in real time
Received signal strength data normalization module (402) is connected, and online data Fusion Module (403) is obtained by data anastomosing algorithm
Target location appears in the probability of reference point, and exports to target location computing module (405)；Read with reference to a geographical position
Enter module (404) to obtain the position coordinates of reference point from characteristic fingerprint storehouse extraction module (300) and be delivered to target location meter
Module (405) is calculated, the module appears in reference point probability by reference to the position coordinates and real-time reception signal strength data of point
The position coordinates of target is calculated, target location output module (500) enters position coordinates corresponding to real-time reception signal intensity
Row output.
The technical program further optimizes, and it is pre- that the four layer depths confidence mixed-media network modules mixed-media (302) includes characteristic fingerprint
Training module (3021), data reconstruction module (3022) and characteristic fingerprint tuning module (3023), the characteristic fingerprint pre-training
Module (3021) is established module (200) with thick fingerprint base and is connected, and the characteristic fingerprint pre-training module (3021) will receive offline
Signal intensity is normalized, and normalized offline received signal strength is trained using four layer depth confidence networks,
Weight between adjacent net network layers and bias matrix are defined as characteristic fingerprint；Characteristic fingerprint pre-training module (3021) and data
Reconstructed module (3022) is connected, and characteristic fingerprint is carried out transposition by the data reconstruction module (3022), is reconstructed by backpropagation
Go out received signal strength data, characteristic fingerprint tuning module (3023) and characteristic fingerprint pre-training module (3021) and data reconstruction
Module (3022) is connected, and the characteristic fingerprint tuning module (3023) is by the offline reception of characteristic fingerprint pre-training module (3021)
Signal intensity and the signal intensity that receives of reconstruct in data reconstruction module (3022) make the difference, if difference is more than predetermined threshold value
When, continue backout feature fingerprint pre-training module (3021) and carry out new characteristic fingerprint training, otherwise output characteristic fingerprint to spy
Levy fingerprint base extraction module (303).
The technical program further optimizes, and the online data Fusion Module (403) is real-time including being sequentially connected
Received signal strength variance computing module (4031), radial function computing module (4032) and location probability computing module
(4033)；The real-time reception signal variance computing module (4031) normalizes module (402) phase with real-time reception signal intensity
Even, the variance of the received signal strength gathered in real time is calculated, and variance is delivered to radial function computing module (4032), it is described
Radial function computing module (4032) is connected with characteristic fingerprint storehouse extraction module (300), radial function computing module (4032) place
The result of reason is delivered to location probability computing module (4033), calculates real-time reception signal strength data and appears in reference point position
The probability put, and final location probability is delivered to target location computing module (405).
The technical program further optimizes, wireless senser in described offline reception signal acquisition module (102)
Sample frequency be arranged to 300Hz.
The technical program further optimizes, and the four layer depths confidence network is generative probabilistic model, by multiple limitations
Boltzmann machine forms four layer depth networks, and the training of characteristic fingerprint is carried out using Greedy algorithm.
The technical program further optimizes, and the RBF is the Gaussian function based on Euclidean distance.
Compared with prior art, the invention has the advantages that：
1. fully rely on intelligent terminal without extra signal transmitting equipment in position fixing process.
2. in off-line phase, more than 100 offline received signal strength datas of collection are only needed at each reference point, input is deeply
Confidence network is spent with regard to characteristic fingerprint can be trained, and is adopted compared to other targeting schemes based on deep learning network in off-line data
Manpower is more saved in terms of collection.
3. using the characteristic fingerprint trained, the high-precision real time position for estimating target can be realized, both meets height
The demand of precision positioning, while also meet the performance of real-time location estimation.
Brief description of the drawings
Fig. 1 is the structure chart of the WiFi indoor locating systems based on deep learning；
Fig. 2 is the structure principle chart that thick fingerprint base is established；
Fig. 3 is characterized the structure principle chart of fingerprint base extraction；
Fig. 4 is the structure principle chart of online data fusion；
Fig. 5 is the structure principle chart of four layer depth confidence networks；
Fig. 6 is the structure principle chart of online data fusion.
Embodiment
To further illustrate each embodiment, the present invention is provided with accompanying drawing.These accompanying drawings are the invention discloses the one of content
Point, it can coordinate the associated description of specification to explain the operation principles of embodiment mainly to illustrate embodiment.Coordinate ginseng
These contents are examined, those of ordinary skill in the art will be understood that other possible embodiments and advantages of the present invention.In figure
Component be not necessarily to scale, and similar element numbers are conventionally used to indicate similar component.
In conjunction with the drawings and specific embodiments, the present invention is further described.
As shown in fig.2, the structure principle chart established for thick fingerprint base.Off-line data acquisition module 100 includes wireless pass
Sensor module 101 and offline received signal strength acquisition module 102.Wireless sensor module 101 is used to scan all access points
XrThe physical address of wireless senser, offline received signal strength acquisition module 102, which is used to gather, is located at reference point YsPlace comes from
Access point XrThe offline of place's wireless senser receives signal strength data, and the signal strength data is not limited to one, can gather
It is numerous.
Thick fingerprint base, which establishes module 200, includes reference point locations logging modle 201, thick fingerprint base generation module 202 and thick
Fingerprint base output module 203.Thick fingerprint base generation module 202 gathers the offline received signal strength acquisition module 102 of reference point
Reference point YsThe offline signal strength data that receives at place records reference point Y with reference point locations logging modle 201sThe position at place is sat
Mark is corresponded, and travels through all reference point YsThick fingerprint base is formed, thick fingerprint includes YsPosition coordinates, offline receive letter
The physical address of number intensity data and wireless senser, all thick fingerprints are delivered to thick fingerprint base output module 203.
The sample frequency of wireless senser could be arranged to 300Hz or so in offline reception signal acquisition module 102, can
Stably collection received signal strength and can gathers more data in a short time.
As shown in fig.3, it is characterized the structure principle chart of fingerprint base extraction.Characteristic fingerprint storehouse extraction module 300 include from
Line reception signal normalization module 301, four layer depth confidence mixed-media network modules mixed-medias 302, characteristic fingerprint extraction module 303 and characteristic fingerprint
Storehouse output module 304.Wherein, receive signal normalization module 301 offline with thick fingerprint base output module 203 to be connected, will be each
Reference point YsThe offline receiving intensity signal normalization at place is between (0,1) so that received signal strength signal can be offline
It is trained in four layer depth confidence mixed-media network modules mixed-medias 302.After whole four layer depths confidence network is completed to train, characteristic fingerprint
The weight that four layer depth confidence network trainings go out is extracted and deposited as characteristic fingerprint by extraction module 303 with bias matrix
Storage.When training all reference point Y with depth confidence network traversersIt is special after locating normalized offline received signal strength data
Sign fingerprint base output module 304 combines all reference point YsThe characteristic fingerprint at place forms characteristic fingerprint storehouse and exported, for online
The real-time positioning that stage carries out target location is used, each reference point YsCharacteristic fingerprint include position coordinates, weight, biasing
Matrix.
Four layer depth confidence networks are generative probabilistic models, and four layer depth networks are formed by multiple limitation Boltzmann machines,
The training of characteristic fingerprint is carried out using Greedy algorithm, the model is as follows：
P (x, h1, h2, h3, h4)=P (x/h1)(h1/h2)P(h2/h3)P(h3/h4)P(h4)
Wherein, x represents input layer, hi(i=1,2,3,4) represents the output variable of four layers of hidden layer respectively, passes through four layers
Training ultimately generates a total probability model, the reconstruct for received signal strength data.The neuron number of each hidden layer
It is that each Internet sets neuron according to the order successively successively decreased according to the dimension of the received signal strength data of collection
Number.The activation primitive of forward direction transmission is sigmoid functions in the neutral net, and gradient descent algorithm is used in reverse transfers,
Carry out the training of characteristic fingerprint.{w1,b1},{w2,b2},{w3,b3And { w4,b4The weight between hidden layer and biasing are represented respectively
Matrix, when the completion of depth confidence network training, extract all weights and bias matrix is stored as characteristic fingerprint, fingerprint
Training is shown below：
Wherein, hi(i=1,2,3,4) represents the output variable of four layers of hidden layer, { w respectivelyi,bj(i=1,2,3,4) expression
Weight and bias matrix corresponding to four layers of hidden layer.
As shown in fig.5, the structure principle chart for four layer depth confidence networks.Four layer depth confidence mixed-media network modules mixed-medias 302 wrap
The module of pre-training containing characteristic fingerprint 3021, data reconstruction module 3022 and characteristic fingerprint tuning module 3023.Wherein characteristic fingerprint
Pre-training module 3021 is connected with offline received signal strength data normalization module 301, to normalized offline reception signal
Intensity data, it is trained using four layer depth confidence networks.Normalized offline received signal strength data is in the first layer depth
Spend in confidence network after the completion of training, the fixed first layer weight trained and bias matrix, i.e. { w1,b1, and as
The input of two layer depth confidence networks.Then the training of the second layer depth confidence network is carried out, obtains the second layer depth confidence net
The weight and bias matrix of network, i.e. { w2,b2, by that analogy, the training of third layer, the 4th layer depth confidence network below is carried out,
The forward direction training of entire depth confidence network is completed, the weight between adjacent net network layers and bias matrix definition are characterized finger
Line, i.e. characteristic fingerprint include { w1,b1},{w2,b2},{w3,b3And { w4,b4}.Characteristic fingerprint and training module 3021 and data weight
Structure module 3022 is connected, and the characteristic fingerprint of pre-training is carried out into transposition, received signal strength data is reconstructed by backpropagation.
Specific method is the backpropagation since the 4th layer, reconstructs third layer network by the transposition of the 4th layer of characteristic fingerprint, tightly
Then the second layer network is reconstructed by the transposition of third layer characteristic fingerprint, by that analogy, reconstructs that original to receive signal strong
Degrees of data, complete data reconstruction module 3022.Characteristic fingerprint tuning module 3023 respectively with characteristic fingerprint pre-training module 3021
It is connected with data reconstructed module 3022, by the offline received signal strength of characteristic fingerprint pre-training module 3021 and data reconstruction mould
The signal intensity that receives reconstructed in block 3022 makes the difference, and judges the size of difference, when difference is more than predetermined threshold value, continues to return
Return characteristic fingerprint pre-training module 3021 and carry out new fingerprint training, go successively to data reconstruction module 3022 and carry out former data
Reconstruct, the difference for entering back into reconstruction signal and original signal judge that, until final difference is less than or equal to predetermined threshold value, completion is special
The training of fingerprint is levied, is exported to characteristic fingerprint storehouse extraction module 303.
As shown in fig.4, being the structure principle chart of online data fusion, online data Fusion Module 400 is included and connect in real time
Receive signal strength data acquisition module 401, real-time reception signal strength data normalization module 402, online data Fusion Module
403rd, module 404 and target location computing module 405 are read in reference point geographical position.Wherein, real-time reception signal intensity gathers
Module 401 gathers the real-time reception signal intensity of access point wireless senser, after completing collection, and then in real-time reception signal
Live signal is normalized to be easy between (0,1) to be handled in depth confidence network in normalization module 402.In line number
It is connected according to Fusion Module 403 with characteristic fingerprint storehouse output module 304 and real-time reception signal strength data normalization module 402,
Target location must be arrived by data anastomosing algorithm and appear in reference point YsProbability output to target location computing module 405；Together
When reference point geographical position read in module 404 reference point Y is obtained from characteristic fingerprint storehouse output module 304sPosition coordinates simultaneously
Target location computing module 405 is delivered to, the module is by reference to point YsPosition coordinates and real-time reception signal strength data
Appear in reference point YsProbability calculation goes out the position coordinates of target, is finally believed real-time reception by target location output module 500
Number position coordinates corresponding to intensity is exported.
As shown in fig.6, being the structure principle chart of online data fusion, the real-time reception of online data Fusion Module 403 is believed
Number intensity variance computing module 4031, radial function computing module 4032 and location probability computing module 4033.Real-time reception is believed
Number variance computing module 4031 is connected with real-time reception signal intensity normalization module 402, calculates the reception gathered in real time and believes
The variance of number intensity, and this variance is delivered to radial function computing module 4032.Other radial function computing module 4032 with
Characteristic fingerprint storehouse output module 304 is connected, and passes through each reference point YsCharacteristic fingerprint carry out received signal strength data joining
Examination point YsThe reconstruct at place, and calculate real-time reception signal intensity vector and the Euclidean distance of reconstructed reception signal intensity, footpath
The variance of the Euclidean distance and real-time reception signal intensity is set up into gauss of distribution function to basic function.RBF
The result calculated is delivered to location probability computing module 4033, calculates real-time reception signal strength data and appears in reference point
YsThe probability of position, and final location probability is delivered to target location computing module 405.
RBF is the Gaussian function based on Euclidean distance, can approach any non-linear relation, in processing data
The complex relationship in portion.Full articulamentum of the RBF as neutral net, centre data are offline received signal strength, are passed through
The characteristic fingerprint acquisition reconstruct trained receives signal intensity, establishes Gaussian function by Euclidean distance, calculates and currently connect
Receive signal intensity and appear in reference point Y corresponding to characteristic fingerprintsProbability.RBF in this method is shown below：
Wherein, P (x/Li) representing prior probability, i.e. current received signal strength appears in reference point YsThe probability x tables of position
Show current received signal strength data,Expression utilizes reference point YsThe characteristic fingerprint reconstruct at place receives signal strength data, σ
Represent the variance of current received signal strength data.
Although specifically showing and describing the present invention with reference to preferred embodiment, those skilled in the art should be bright
In vain, do not departing from the spirit and scope of the present invention that appended claims are limited, in the form and details can be right
The present invention makes a variety of changes, and is protection scope of the present invention.
Claims (10)
- A kind of 1. WiFi indoor locating systems based on deep learning, it is characterised in that：It includes the off-line data being sequentially connected Acquisition module (100), thick fingerprint base establish module (200), characteristic fingerprint storehouse extraction module (300), online data Fusion Module (400) and target location output module (500), wherein,The off-line data acquisition module (100) is used to obtaining the physical address information of access point, collection at reference point from Line received signal strength data, and these data are transferred to thick fingerprint base and establish module (200)；The offline received signal strength data that the thick fingerprint base is established at all reference points of module (200) traversal so that offline Received signal strength and the position coordinates of reference point, which correspond, generates thick fingerprint base；Receive signal strength data offline at characteristic fingerprint storehouse extraction module (300) normalization reference point, be input to Four layer depth confidence networks are trained, and four layer depth confidence networks output weight and bias matrix are stored as characteristic fingerprint；The online data Fusion Module (400) gathers received signal strength data and it is normalized in real time, then The received signal strength data gathered in real time is reconstructed using the characteristic fingerprint at each reference point, by the receiving letter of reconstruct Number intensity data and the former data input that gathers in real time calculate diversity factor between the two, traversal is all to RBF Reference point, probability of the current received signal strength data at each reference point is calculated, finally by reference point position The weight calculation put goes out geographical position corresponding to real-time reception signal strength data；Described target location output module (500) will be exported based on the target location that data anastomosing algorithm estimates, complete The positioning of paired target location.
- 2. the WiFi indoor locating systems according to claim 1 based on deep learning, it is characterised in that the offline number Include wireless sensor module (101) and offline received signal strength acquisition module (102) according to acquisition module (100), it is wireless to pass Sensor module (101) is used for the physical address for scanning all access point wireless sensers, offline received signal strength acquisition module (102) come from the offline of access point wireless senser at reference point for collection and receive signal strength data.
- 3. the WiFi indoor locating systems according to claim 2 based on deep learning, it is characterised in that the thick fingerprint It is defeated including reference point locations logging modle (201), thick fingerprint base generation module (202) and thick fingerprint base that module (200) is established in storehouse Go out module (203)；Thick fingerprint base is established module (200) and will gathered by the offline received signal strength acquisition module (102) of reference point The offline position seat for receiving signal strength data and reference point being recorded by reference point locations logging modle (201) at reference point Mark and corresponded, travel through all reference points and form thick fingerprint base, be delivered to thick fingerprint base output module (203).
- 4. the WiFi indoor locating systems according to claim 1 based on deep learning, it is characterised in that the feature refers to Line storehouse extraction module (300) includes offline reception signal normalization module (301), four layer depth confidence mixed-media network modules mixed-medias (302), spy Levy fingerprint extraction module (303) and characteristic fingerprint storehouse output module (304)；It is offline receive signal normalization module (301) with it is thick Fingerprint base is established module (200) and is connected, and by the offline receiving intensity signal normalization at each reference point between (0,1), incites somebody to action Normalized offline received signal strength signal is trained in four layer depth confidence mixed-media network modules mixed-medias (302), and characteristic fingerprint carries Modulus block (303) extracts the weight in four layer depth confidence mixed-media network modules mixed-medias (302) and bias matrix as characteristic fingerprint Stored, the characteristic fingerprint that characteristic fingerprint storehouse output module (304) is combined at all reference points forms characteristic fingerprint storehouse and carried out Output.
- 5. the WiFi indoor locating systems according to claim 1 based on deep learning, it is characterised in that described in line number Include real-time reception signal strength data acquisition module (401), real-time reception signal strength data normalizing according to Fusion Module (400) Change module (402), online data Fusion Module (403), module (404) is read in reference point geographical position and target location calculates mould Block (405)；The real-time reception signal of real-time reception signal intensity acquisition module (401) the collection access point wireless senser is strong Degree, between live signal is normalized to (0,1) by real-time reception signal normalization module (402), online data Fusion Module (403) it is connected with characteristic fingerprint storehouse extraction module (300) and real-time reception signal strength data normalization module (402), online Data fusion module (403) must arrive the probability that target location appears in reference point by data anastomosing algorithm, and export to target Position computation module (405)；Module (404) is read in from characteristic fingerprint storehouse extraction module (300) with reference to a geographical position Obtain reference point position coordinates simultaneously be delivered to target location computing module (405), the module by reference to point position coordinates And real-time reception signal strength data appears in the position coordinates that reference point probability calculation goes out target, target location output module (500) position coordinates corresponding to real-time reception signal intensity is exported.
- 6. the WiFi indoor locating systems according to claim 4 based on deep learning, it is characterised in that four layer depth Degree confidence mixed-media network modules mixed-media (302) includes characteristic fingerprint pre-training module (3021), data reconstruction module (3022) and characteristic fingerprint Tuning module (3023), the characteristic fingerprint pre-training module (3021) are established module (200) with thick fingerprint base and are connected, the spy Offline received signal strength is normalized sign fingerprint pre-training module (3021), and using four layer depth confidence networks to returning The one offline received signal strength changed is trained, and the weight between adjacent net network layers and bias matrix definition are characterized into finger Line；Characteristic fingerprint pre-training module (3021) is connected with data reconstruction module (3022), and the data reconstruction module (3022) will Characteristic fingerprint carries out transposition, and received signal strength data is reconstructed by backpropagation, characteristic fingerprint tuning module (3023) with Characteristic fingerprint pre-training module (3021) is connected with data reconstructed module (3022), and the characteristic fingerprint tuning module (3023) will The offline received signal strength of characteristic fingerprint pre-training module (3021) and the receiving letter of reconstruct in data reconstruction module (3022) Number intensity makes the difference, if difference is more than predetermined threshold value, continue backout feature fingerprint pre-training module (3021) carry out it is new Characteristic fingerprint is trained, otherwise output characteristic fingerprint to characteristic fingerprint storehouse extraction module (303).
- 7. the WiFi indoor locating systems according to claim 5 based on deep learning, it is characterised in that described in line number Real-time reception signal intensity variance computing module (4031), the radial function for including being sequentially connected according to Fusion Module (403) calculate Module (4032) and location probability computing module (4033)；The real-time reception signal variance computing module (4031) with connecing in real time Receive signal intensity normalization module (402) to be connected, calculate the variance of the received signal strength gathered in real time, and variance is delivered to Radial function computing module (4032), the radial function computing module (4032) and characteristic fingerprint storehouse extraction module (300) phase Even, the result of radial function computing module (4032) processing is delivered to location probability computing module (4033), calculates and connects in real time Receive signal strength data and appear in the probability of reference point locations, and final location probability is delivered to target location computing module (405)。
- 8. the WiFi indoor locating systems according to claim 2 based on deep learning, it is characterised in that described is offline The sample frequency of wireless senser is arranged to 300Hz in reception signal acquisition module (102).
- 9. the WiFi indoor locating systems according to claim 1 based on deep learning, it is characterised in that four layer depth It is generative probabilistic model to spend confidence network, four layer depth networks is formed by multiple limitation Boltzmann machines, using Greedy algorithm Carry out the training of characteristic fingerprint.
- 10. the WiFi indoor locating systems according to claim 1 based on deep learning, it is characterised in that the radial direction Basic function is the Gaussian function based on Euclidean distance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710833040.8A CN107529222B (en) | 2017-09-15 | 2017-09-15 | WiFi indoor positioning system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710833040.8A CN107529222B (en) | 2017-09-15 | 2017-09-15 | WiFi indoor positioning system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107529222A true CN107529222A (en) | 2017-12-29 |
| CN107529222B CN107529222B (en) | 2020-11-06 |
Family
ID=60736896
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710833040.8A Active CN107529222B (en) | 2017-09-15 | 2017-09-15 | WiFi indoor positioning system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107529222B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108363086A (en) * | 2018-02-26 | 2018-08-03 | 成都步速者科技股份有限公司 | Indoor navigation method, device, server and storage medium |
| CN108462992A (en) * | 2018-03-05 | 2018-08-28 | 中山大学 | A kind of indoor orientation method based on super-resolution rebuilding Wi-Fi fingerprint maps |
| CN108566620A (en) * | 2018-04-18 | 2018-09-21 | 南京小木马科技有限公司 | A kind of indoor orientation method based on WIFI |
| CN108769969A (en) * | 2018-06-20 | 2018-11-06 | 吉林大学 | A kind of RFID indoor orientation methods based on depth confidence network |
| CN108961450A (en) * | 2018-06-29 | 2018-12-07 | 夏烬楚 | A kind of attendance system and method based on channel state information |
| CN110333484A (en) * | 2019-07-15 | 2019-10-15 | 桂林电子科技大学 | The room area grade localization method with analysis is known based on environmental background phonoreception |
| CN110418407A (en) * | 2019-08-27 | 2019-11-05 | 成都市东信德科技有限公司 | Exception luggage bluetooth localization method neural network based and its system |
| CN110823221A (en) * | 2018-08-13 | 2020-02-21 | 哈尔滨海能达科技有限公司 | Indoor positioning system, method and device |
| CN111090090A (en) * | 2019-12-11 | 2020-05-01 | 金华航大北斗应用技术有限公司 | Method for constructing feature fingerprint database in indoor positioning system |
| EP3736596A1 (en) * | 2019-05-06 | 2020-11-11 | Siemens Healthcare GmbH | Add-on module for a device, server device, positioning method, computer program and corresponding storage medium |
| CN111935628A (en) * | 2020-07-28 | 2020-11-13 | 河南大学 | Wi-Fi positioning method and device based on position fingerprint |
| CN113253725A (en) * | 2021-05-11 | 2021-08-13 | 北京京东乾石科技有限公司 | Robot path planning method and device, storage medium and electronic equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103796304A (en) * | 2014-01-15 | 2014-05-14 | 内蒙古科技大学 | Coal mine underground positioning method based on virtual training set and Markov chain |
| CN103945533A (en) * | 2014-05-15 | 2014-07-23 | 济南嘉科电子技术有限公司 | Big data based wireless real-time position positioning method |
| CN105101408A (en) * | 2015-07-23 | 2015-11-25 | 常熟理工学院 | Indoor positioning method based on distributed AP selection strategy |
| CN106793070A (en) * | 2016-11-28 | 2017-05-31 | 上海斐讯数据通信技术有限公司 | A kind of WiFi localization methods and server based on reinforcement deep neural network |
| CN107037399A (en) * | 2017-05-10 | 2017-08-11 | 重庆大学 | A kind of Wi Fi indoor orientation methods based on deep learning |
| US10051475B2 (en) * | 2015-09-28 | 2018-08-14 | Department 13, Inc. | Unmanned aerial vehicle intrusion detection and countermeasures |
-
2017
- 2017-09-15 CN CN201710833040.8A patent/CN107529222B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103796304A (en) * | 2014-01-15 | 2014-05-14 | 内蒙古科技大学 | Coal mine underground positioning method based on virtual training set and Markov chain |
| CN103945533A (en) * | 2014-05-15 | 2014-07-23 | 济南嘉科电子技术有限公司 | Big data based wireless real-time position positioning method |
| CN105101408A (en) * | 2015-07-23 | 2015-11-25 | 常熟理工学院 | Indoor positioning method based on distributed AP selection strategy |
| US10051475B2 (en) * | 2015-09-28 | 2018-08-14 | Department 13, Inc. | Unmanned aerial vehicle intrusion detection and countermeasures |
| CN106793070A (en) * | 2016-11-28 | 2017-05-31 | 上海斐讯数据通信技术有限公司 | A kind of WiFi localization methods and server based on reinforcement deep neural network |
| CN107037399A (en) * | 2017-05-10 | 2017-08-11 | 重庆大学 | A kind of Wi Fi indoor orientation methods based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 覃玉清: "基于深度学习的WiFi定位算法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108363086A (en) * | 2018-02-26 | 2018-08-03 | 成都步速者科技股份有限公司 | Indoor navigation method, device, server and storage medium |
| CN108462992A (en) * | 2018-03-05 | 2018-08-28 | 中山大学 | A kind of indoor orientation method based on super-resolution rebuilding Wi-Fi fingerprint maps |
| CN108462992B (en) * | 2018-03-05 | 2021-03-19 | 中山大学 | Indoor positioning method based on super-resolution reconstruction Wi-Fi fingerprint map |
| CN108566620B (en) * | 2018-04-18 | 2021-01-05 | 南京市木阿码网络信息科技有限公司 | Indoor positioning method based on WIFI |
| CN108566620A (en) * | 2018-04-18 | 2018-09-21 | 南京小木马科技有限公司 | A kind of indoor orientation method based on WIFI |
| CN108769969A (en) * | 2018-06-20 | 2018-11-06 | 吉林大学 | A kind of RFID indoor orientation methods based on depth confidence network |
| CN108769969B (en) * | 2018-06-20 | 2021-10-15 | 吉林大学 | RFID indoor positioning method based on deep belief network |
| CN108961450A (en) * | 2018-06-29 | 2018-12-07 | 夏烬楚 | A kind of attendance system and method based on channel state information |
| CN110823221A (en) * | 2018-08-13 | 2020-02-21 | 哈尔滨海能达科技有限公司 | Indoor positioning system, method and device |
| EP3736596A1 (en) * | 2019-05-06 | 2020-11-11 | Siemens Healthcare GmbH | Add-on module for a device, server device, positioning method, computer program and corresponding storage medium |
| US11774568B2 (en) | 2019-05-06 | 2023-10-03 | Siemens Healthcare Gmbh | Add-on module for a device, server unit, localization method, computer program, and corresponding storage medium |
| CN110333484B (en) * | 2019-07-15 | 2021-04-13 | 桂林电子科技大学 | Indoor area level positioning method based on environmental background sound perception and analysis |
| CN110333484A (en) * | 2019-07-15 | 2019-10-15 | 桂林电子科技大学 | The room area grade localization method with analysis is known based on environmental background phonoreception |
| CN110418407A (en) * | 2019-08-27 | 2019-11-05 | 成都市东信德科技有限公司 | Exception luggage bluetooth localization method neural network based and its system |
| CN111090090A (en) * | 2019-12-11 | 2020-05-01 | 金华航大北斗应用技术有限公司 | Method for constructing feature fingerprint database in indoor positioning system |
| CN111090090B (en) * | 2019-12-11 | 2022-05-27 | 金华航大北斗应用技术有限公司 | Method for constructing feature fingerprint database in indoor positioning system |
| CN111935628A (en) * | 2020-07-28 | 2020-11-13 | 河南大学 | Wi-Fi positioning method and device based on position fingerprint |
| CN111935628B (en) * | 2020-07-28 | 2022-06-28 | 河南大学 | Wi-Fi positioning method and device based on position fingerprint |
| CN113253725A (en) * | 2021-05-11 | 2021-08-13 | 北京京东乾石科技有限公司 | Robot path planning method and device, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107529222B (en) | 2020-11-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107529222A (en) | A kind of WiFi indoor locating systems based on deep learning | |
| CN104615983B (en) | Activity recognition method based on recurrent neural network and human skeleton motion sequence | |
| WO2017133009A1 (en) | Method for positioning human joint using depth image of convolutional neural network | |
| CN109115205A (en) | A kind of indoor fingerprint positioning method and system based on geomagnetic sensor array | |
| CN103945533A (en) | Big data based wireless real-time position positioning method | |
| CN110765882B (en) | Video tag determination method, device, server and storage medium | |
| CN109165540A (en) | A kind of pedestrian's searching method and device based on priori candidate frame selection strategy | |
| CN110536245A (en) | A kind of indoor wireless positioning method and system based on deep learning | |
| CN109101938A (en) | A kind of multi-tag age estimation method based on convolutional neural networks | |
| CN111368759B (en) | Monocular vision-based mobile robot semantic map construction system | |
| CN108961330A (en) | The long measuring method of pig body and system based on image | |
| CN111586605B (en) | KNN indoor target positioning method based on adjacent weighted self-adaptive k value | |
| CN110222780A (en) | Object detecting method, device, equipment and storage medium | |
| CN112419326B (en) | Image segmentation data processing method, device, equipment and storage medium | |
| WO2022184124A1 (en) | Physiological electrical signal classification and processing method and apparatus, computer device, and storage medium | |
| CN108629295A (en) | Corner terrestrial reference identification model training method, the recognition methods of corner terrestrial reference and device | |
| CN110072183A (en) | Passive type location fingerprint base construction method based on intelligent perception | |
| CN111160111A (en) | Human body key point detection method based on deep learning | |
| CN113518307A (en) | Position fingerprint positioning method based on CSI and crowdsourcing migration self-calibration updating | |
| CN110096991A (en) | A kind of sign Language Recognition Method based on convolutional neural networks | |
| Chen et al. | A wifi indoor localization method based on dilated cnn and support vector regression | |
| CN112749663A (en) | Agricultural fruit maturity detection system based on Internet of things and CCNN model | |
| WO2022242018A1 (en) | Indoor target positioning method based on improved cnn model | |
| CN114219853A (en) | Multi-person three-dimensional attitude estimation method based on wireless signals | |
| CN112333652B (en) | WLAN indoor positioning method and device and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |